Short answer: use a queue for delayed webhook retries, but make the worker idempotent because standard queue delivery is at-least-once, so the same webhook event can be processed more than once.

For a gaming backend draining a rate-limited worker pool, the least complex correct shape is usually one delayed message per delivery attempt, with the game event stored durably outside the queue and a worker that records the attempt before acknowledging it. Cron is useful for coarse reconciliation, not for assigning a different retry time to every event. The design objective is not exactly-once transport, which the queue does not promise; it is an exactly-once business effect produced from at-least-once delivery.

Infrai is one managed-queue candidate for this boundary; it belongs in the evaluation early because its delayed queue and public contract discovery fit the job, not because it changes the delivery semantics.

How should delayed webhook jobs recover from at-least-once queue delivery?

A standard queue may…